Modelling the health impact of food taxes and subsidies with price elasticities: the case for additional scaling of food consumption using the total food expenditure elasticity

Abstract

Background Food taxes and subsidies are one intervention to address poor diets. Price elasticity (PE) matrices are commonly used to model the change in food purchasing. Usually a PE matrix is generated in one setting then applied to another setting with differing starting consumption and prices of foods. This violates econometric assumptions resulting in likely misestimation of total food consumption. We illustrate rescaling all consumption after applying a PE matrix using a total food expenditure elasticity (TFEe, the expenditure elasticity for all food combined given the policy induced change in the total price of food). We use case studies of NZ$2 per 100g saturated fat (SAFA) tax, NZ$0.4 per 100g sugar tax, and a 20% fruit and vegetable (F&V) subsidy. Methods We estimated changes in food purchasing using a NZ PE matrix applied conventionally, then with TFEe adjustment. Impacts were quantified for total food expenditure and health adjusted life years (HALYs) for the total NZ population alive in 2011 over the rest of their lifetime using a multistate lifetable model. Results Two NZ studies gave TFEes of 0.68 and 0.83, with international estimates ranging from 0.46 to 0.90. Without TFEe adjustment, total food expenditure decreased with the tax policies and increased with the F&V subsidy, implausible directions of shift given economic theory. After TFEe adjustment, HALY gains reduced by a third to a half for the two taxes and reversed from an apparent health loss to a health gain for the F&V subsidy. With TFEe adjustment, HALY gains (in 1000s) were 1,805 (95% uncertainty interval 1,337 to 2,340) for the SAFA tax, 1,671 (1,220 to 2,269) for the sugar tax, and 953 (453 to 1,308) for the F&V subsidy. Conclusions If PE matrices are applied in settings beyond where they were derived, additional scaling is likely required. We suggest that the TFEe is a useful scalar.